List your company

Policy-adaptive safety guardrailing on POLICYSHIFTBENCH (Adaptive split)

86.8F1 ScorePolicyShiftGuard-7B
24.440.656.873Jul 7, 2026

Evaluation results

MethodLinks
PolicyShiftGuard-7BParameter Scale=7B2026.0786.869.9
PolicyShiftGuard-3BParameter Scale=3B2026.0785.662.4
GuardReasoner-VL-3BParameter Scale=3B2026.0763.651.8
SafeGuard-VL-RL-7BParameter Scale=7B2026.0762.552.5
GuardReasoner-VL-7BParameter Scale=7B2026.0761.548.2
QwenGuard-7BParameter Scale=7B2026.073938.7
LlamaGuard4-12BParameter Scale=12B2026.0729.219.1
Qwen2.5-VL-7BParameter Scale=7B2026.072932.9
ShieldGemma2-4BParameter Scale=4B2026.0726.836.8